For many schools, the arrival of generative AI has been treated as an academic honesty crisis. Teachers worry that students are using AI to write essays, complete homework, generate projects, solve problems, and polish responses beyond their actual level of understanding. These concerns are valid. But they are also only the surface of the issue.
The deeper question is not simply, “How do we stop students from using AI?”
The stronger question is:
“What kinds of assessment still reveal real learning in a world where polished answers can be generated instantly?”
This is where the conversation becomes more interesting, and more urgent. AI is not only challenging assessment security. It is exposing weaknesses that were already present in many assessment systems: over-reliance on final products, generic tasks, take-home essays without process evidence, formulaic rubrics, and assignments that reward fluent output more than original thinking.
In this sense, AI has not destroyed assessment. It has revealed the need to redesign it.
The problem is not AI. The problem is weak evidence of learning.
For years, many schools have assessed students through products that can be completed away from the teacher’s view: essays, slideshows, reports, posters, summaries, and research projects. These tasks can be meaningful when designed well. But when they are too generic, too disconnected from classroom learning, or too focused on final submission, they become vulnerable.
AI simply makes this vulnerability impossible to ignore.
A student can now submit work that is grammatically polished, structurally coherent, and superficially analytical without having gone through the necessary struggle of thinking. This creates what many educators are beginning to recognize as a dangerous gap between performance and learning.
The OECD’s Digital Education Outlook 2026 argues that generative AI can support learning when guided by clear teaching goals and educational principles, but it also warns that AI use must not replace students’ cognitive effort. (OECD)
That distinction matters. AI can help students brainstorm, clarify, question, translate, organize, and revise. But if AI does the intellectual heavy lifting, the student may end up with a better product and weaker learning.
This is the central assessment challenge of the AI age:
A polished product is no longer enough evidence that learning has happened.
From academic integrity to assessment integrity
Schools often respond to AI through academic integrity policies. That is necessary, but not sufficient.
Academic integrity asks:
Did the student act honestly?
Assessment integrity asks:
Does this task actually show what the student knows, understands, and can do?
These are connected but not identical.
A student may follow the rules and still complete a poorly designed task that does not reveal much learning. Another student may misuse AI because the task itself invites outsourcing: “Write a five-paragraph essay about climate change,” “Make a presentation about Shakespeare,” or “Summarize this historical event.”
These tasks are not automatically bad, but they become weak when they are detached from process, voice, context, and classroom evidence.
The International Baccalaureate has updated its academic integrity guidance to include the use of artificial intelligence tools, emphasizing ethical use rather than pretending AI can be removed from learning altogether. (International Baccalaureate®) This is an important shift. The future of assessment is not about pretending students will never use AI. It is about designing learning and assessment so that students must demonstrate understanding in ways that cannot be reduced to generated output.
The question becomes:
What evidence would convince us that this student has genuinely learned?
The death of the “one-and-done” assignment
The traditional model of assessment often looks like this:
Teacher gives task.
Student works independently.
Student submits final product.
Teacher grades final product.
In the AI age, this model is too fragile.
The future of assessment needs to look more like this:
Teacher introduces a meaningful problem or inquiry.
Students build background knowledge.
Students discuss, question, draft, test, revise, and reflect.
Teacher gathers evidence along the way.
Students explain their choices.
Final product is assessed alongside process, thinking, and transfer.
This does not mean every assessment must become complicated. It means that assessment must become more visible.
Teachers need to see more than the answer. They need to see the thinking behind the answer.
What should assessment redesign look like?
A strong AI-era assessment system should include several types of evidence.
1. Process evidence
Students should be expected to show how their thinking developed. This may include notes, outlines, drafts, annotations, research logs, planning documents, feedback responses, version history, and reflections.
The point is not to create more paperwork. The point is to make learning visible.
For example, instead of asking students only to submit a final essay, we might ask for:
A research question
A source evaluation
A planning outline
One paragraph drafted in class
Peer or teacher feedback
A revised paragraph
A short reflection explaining what changed and why
The final essay
This makes it much harder for students to outsource the entire task. More importantly, it teaches students that good work is built through revision.
2. Oral defense and student voice
One of the simplest and most powerful tools in AI-era assessment is conversation.
After submitting a piece of work, students can be asked to explain:
Why did you choose this argument?
Which source influenced your thinking most?
What changed between your first and final draft?
Which part are you most confident about?
Which part was difficult?
How would you improve this if you had more time?
What would you do differently in another context?
A short oral defense does not need to be intimidating. It can be a three-minute conference, a recorded reflection, a small-group discussion, or a teacher-student check-in.
The purpose is not to catch students. The purpose is to reconnect assessment with ownership.
When students can explain their thinking, the work becomes more authentic.
3. In-class checkpoints
If the whole task happens at home, the teacher sees only the final outcome. In-class checkpoints allow the teacher to observe the student’s actual level of independence.
This could include:
writing one paragraph in class,
solving a similar problem without support,
annotating a text live,
explaining a concept to a partner,
creating a mind map from memory,
responding to an unseen prompt,
or applying the same skill to a new example.
These moments do not need to replace larger projects. They strengthen them.
A take-home project paired with in-class evidence gives a more reliable picture of learning.
4. Authentic and contextualized tasks
AI is very good at generic responses. Therefore, assessment tasks need to become less generic.
Instead of asking:
“Write an essay about persuasive techniques.”
We might ask:
“Analyze how two climate awareness campaigns use persuasive techniques differently for different audiences. Then create a short campaign for our school community and justify your choices using AFOREST.”
Instead of asking:
“Explain the theme of identity in the novel.”
We might ask:
“Choose one moment where the character’s identity is challenged. Explain how the author uses language, point of view, and conflict to shape the reader’s understanding. Then connect this moment to one discussion we had in class.”
The more a task is connected to classroom learning, local context, student choice, and specific texts or experiences, the harder it is to outsource meaningfully.
5. Transfer tasks
One of the strongest signs of understanding is transfer: the ability to apply learning in a new but related situation.
If students have learned how to analyze persuasive techniques in advertisements, can they analyze a political speech, a school campaign, a charity poster, or a social media video?
If students have learned about characterization in one story, can they apply the same concepts to a new extract?
If students have learned a mathematical method, can they decide when it is useful and when it is not?
AI may generate an answer, but transfer requires flexible understanding. That is what schools should assess more deliberately.
The role of AI should be declared, not hidden
One mistake schools can make is forcing AI use underground. If students believe that any use of AI is automatically cheating, they may hide it rather than learn how to use it ethically.
A more mature approach is to require transparency.
Students can be asked to include an AI use statement:
Did you use AI?
For what purpose?
What prompts did you use?
What did you accept, reject, or revise?
How did you verify the information?
Where is your own thinking visible?
This aligns with the broader direction of responsible AI guidance. UNESCO’s guidance on generative AI in education emphasizes the need for human-centered use, institutional policies, human agency, and protection from risks such as privacy concerns and uncritical dependence on tools. (UNESCO)
The goal is not to normalize shortcuts. The goal is to teach ethical judgment.
In the same way that students learn to cite sources, they now need to learn how to acknowledge AI assistance.
Assessment should classify AI use by purpose
Not all AI use is the same.
There is a major difference between:
Using AI to generate an entire essay
Using AI to brainstorm possible research questions
Using AI to explain a difficult concept
Using AI to improve grammar
Using AI to translate a word or phrase
Using AI to create a study quiz
Using AI to challenge an argument
Using AI to summarize a source the student has not read
Schools need more nuanced language than “allowed” or “not allowed.”
A useful framework could be:
Green use: generally acceptable with transparency
Brainstorming ideas
Creating study questions
Checking grammar
Generating practice examples
Asking for explanations of difficult concepts
Using AI as a tutor after attempting the task
Yellow use: acceptable only with teacher guidance
Revising structure
Getting feedback on an argument
Using AI to compare ideas
Translating larger sections
Using AI to organize research notes
Generating counterarguments
Red use: not acceptable
Submitting AI-generated work as one’s own
Using AI to write assessed paragraphs or essays
Inventing or fabricating sources
Using AI during restricted assessments
Using AI to avoid reading, thinking, or demonstrating required skills
This kind of framework helps students understand that the issue is not the tool itself. The issue is whether the tool supports or replaces learning.
Teachers also need assessment literacy in the AI age
It is unfair to expect teachers to solve this alone. AI-era assessment requires professional development, collaborative planning, leadership support, and time.
Teachers need space to ask:
Which of our assessments are still reliable?
Which tasks are too generic?
Where can we add process evidence?
Where can we use oral explanation?
How do we make expectations clear to students?
How do we avoid increasing workload?
How do we assess fairly when students have unequal access to AI tools?
How do we teach ethical use rather than only punish misuse?
Assessment redesign should not become another burden placed on teachers. It should be a school-wide conversation.
The OECD notes that generative AI is reshaping education beyond teaching and learning because it is widely accessible and often used outside institutional control. (OECD Events) That means schools cannot rely only on surveillance or detection. They need shared principles.
AI detection is not a sustainable assessment strategy
Many schools have turned to AI detectors. But detection tools are unreliable, easy to challenge, and potentially harmful when they falsely accuse students. Even when detection tools improve, they do not solve the deeper problem.
A detector may claim whether a text looks AI-generated. It cannot tell us what a student understands.
The better approach is not detection-first. It is design-first.
Instead of asking, “Can we detect cheating?” we should ask:
“Have we designed the task so that the student must show thinking, process, and ownership?”
This is a healthier question. It moves schools from suspicion to pedagogy.
What this means for IB schools
For IB schools, AI-era assessment redesign should feel familiar rather than frightening.
The IB already values inquiry, conceptual understanding, reflection, transfer, academic integrity, criterion-related assessment, and student agency. These are exactly the principles schools need now.
In the MYP, for example, assessment redesign can be strongly connected to:
ATL skills
process journals
criterion-related rubrics
oral communication
reflection
interdisciplinary learning
community and personal projects
conceptual understanding
global contexts
inquiry questions
authentic performance tasks
The challenge is not to abandon IB principles. The challenge is to apply them more deliberately.
A weak assessment asks students to produce an answer.
A strong IB assessment asks students to demonstrate understanding, justify choices, reflect on process, and transfer learning.
That is precisely the direction schools need in the age of AI.
A practical model: the AI-resilient assessment cycle
Schools can use a simple cycle when redesigning assessments.
Step 1: Clarify the learning
What do we actually want students to know, understand, and be able to do?
If the objective is analysis, do not assess summary.
If the objective is argument, do not assess decoration.
If the objective is conceptual understanding, do not assess copied definitions.
Step 2: Identify the risk
Which parts of this task could be easily outsourced to AI?
Could AI write the whole response?
Could AI generate the research?
Could AI complete the reflection?
Could AI create the product without the student learning?
Step 3: Add visible thinking
Where will students show their process?
Drafts
Annotations
Planning
Conferences
Research logs
Peer feedback
Teacher checkpoints
Reflections
Step 4: Add student voice
How will students explain their work?
Oral defense
Recorded explanation
Teacher conference
Peer discussion
Presentation with questioning
Reflection interview
Step 5: Add transfer
How will students apply the learning to a new situation?
Unseen text
New problem
New audience
New context
Comparative task
Real-world scenario
Step 6: Clarify AI expectations
What AI use is allowed, limited, or prohibited?
Students should know this before they begin, not after they submit.
The future of assessment is more human, not less
There is an irony in this moment. AI may push education to become more human.
If machines can generate fluent answers, then schools must pay closer attention to the things that are deeply human: curiosity, judgment, voice, creativity, ethical reasoning, personal connection, struggle, revision, dialogue, and transfer.
Assessment should not only measure what students can produce. It should reveal how they think.
This is not a small change. It requires courage from schools. It requires leaders to protect time for teachers. It requires teachers to redesign tasks thoughtfully. It requires students to understand that learning is not the same as producing.
But the opportunity is powerful.
AI can either make assessment more superficial, or it can push us to make assessment more meaningful.
The choice depends on how schools respond.
If we respond with fear, we will chase detection tools and police students endlessly.
If we respond with clarity, we can build assessment systems that are more authentic, more equitable, and more focused on real learning.
The future of assessment is not about controlling every tool students use.
It is about designing learning experiences where students must show who they are as thinkers.
And perhaps that is the redesign education needed all along.